Fuzzy min-max neural networks: from classification to regression

摘要

In this paper we show two new learning algorithms for a fuzzy min-max neural network. The top down fuzzy min-max (TDFMM) algorithm modifies the classic Simpson's learning algorithm overcoming its main difficulties: the dependence on the presentation order of the patterns and the poor resolutive adaptation to the characteristics of input space. The top down fuzzy min-max regressor (TDFMMR) algorithm extends our neural network to solve regression problems by using a hybrid fuzzy classifier and a gradient descent algorithm.